So I’ve been trying to use AI in a lot of different situations to best understand where it shines. Ultimately, it was a long uphill battle, but it’s now become clear that it’s good enough for the part time enthusiast:

• Home assistant hacker
• Home lab wannabe
• iOS shortcuts newbie
• Blog CMS second timer

If you’re a casual tech tinkerer, AI can make you productive in minutes instead of weeks. Just be careful — it’s not always right.

Why it works

As a part time enthusiast in all of the themes above, I don’t get to work on any of these platforms long enough to develop a full understanding of them. If I dabble in home assistant and then check back in 6 months later, I’ve forgotten every single thing I learned. I can’t remember what entities are called, I can’t remember what plug-ins I installed last time. I’m starting from scratch all over again… and it’s usually a long slow process to ramp back up.

The advantage is that AI can put all that knowledge at your fingertips. In the Home Assistant scenario, I can pass it all my config and log files, and ask it to help me set up the new thing I want to configure. It will spoon-feed me basic instructions, and I’m able to be “proficient” almost immediately.

Detailed example

Another area in my life where I can waste an enormous amount of time is my blog. I chose Ghost CMS because it’s very very easy to use, and integrates perfectly with my writing app of choice: Ulysses. But the downside of the minimalist UI is that it doesn’t have a full WYSIWYG theme editor, and when I want to make customizations, I need to fiddle with the code.

The first time I set this up, I just hired someone off Fiverr. They took an existing theme and modified it, and updated the layout of the blog page. This cost me about $250 and took the guy almost two weeks, as I’m assuming I wasn’t the highest priority customer.

Fast forward to today, I just downloaded the theme files, and asked ChatGPT how I could go about adding in my own tag navigation at the top of my theme page. I even gave the AI screenshots of other blogs that I liked, and asked it what vocabulary to use in order to replicate the same functionality.

ChatGPT talked me through what the developer had done, how he had structured the files, and where I needed to make changes. When I ran into roadblocks, I was able to paste error codes back into a prompt. When I got undesired outcomes, I was able to pass in a screenshot and it would immediately identify the problem

This felt like gaining a super power. I’ll never launch a Ghost site building agency, I’ll never win any design awards, but ChatGPT gets me to the point where I can get the job done and focus on other things.

Where it goes wrong

Sometimes ChatGPT will confidently guide you down the wrong rabbit hole. In one instance, I asked ChatGPT if it could write an iOS shortcut that could check my ghost CMS and tell me the count of published posts. I created the shortcuts, got the API key, ran the script and got an error back. When I passed this error back to ChatGPT, it only then realized that this was not a possibility at all, the Ghost API doesn’t provide access to the “scheduled” posts.

In another situation, I was having Roo Code write me a script that helped parse through a CSV of e-mail exports. The exports were long threads of e-mails, and I wanted to analyze them e-mail by e-mail, so I needed the script to help me break up the text. It wrote a v1 of the script that worked reasonably well, then it built a comparison script to help check how accurate the initial script was, and then did a couple rounds of modifications.

The comparison script kept telling us that the accuracy was improving, so I thought we were making progress, until I checked by hand. What I found was that the data had been corrupted, and basically the comparison script had broken. All the improvements that I thought it had been making actually had corrupted the file, and I had not saved a backup when the script was partially working.

In the end, it just helps to be aware of these limitations. If you know what to pay attention to and when, you can double-check the AI’s work and catch it before it goes too far in the wrong direction.

Final Thoughts

AI is now at the point where being a part time enthusiast in any technical side hobby is now an option. To reach the same level of skill, I’d probably need to be using these tools for hours per week every single week. Trying to remember syntax, the operating environment, and past configuration decisions was just too much. Now I’m thrilled that I can be an amateur home automator, an amateur blogger, and an amateur python script writer with very little effort.